Predicting Metro Ridership in Dubai: Analyzing Seasonal Trends with SARIMA, Holt-Winters, and LSTM

This paper investigates using three forecasting models-SARIMA, Holt-Winters, and LSTM-to predict ridership demand for the Dubai Metro. A dataset, covering 2016 to August 2024, was preprocessed to remove disruptions from major events and adjusted using a multiplicative seasonality model to better reflect underlying ridership trends. The objective was to generate accurate demand forecasts to support operational planning and resource allocation. The data was aggregated for the entire system monthly to facilitate strategic decisions related to staffing and capacity planning, particularly for medium-term forecasts of up to 2 years. The models were evaluated using key metrics, including MSE, RMSE, MAE, and MAPE. SARIMA emerged as the most accurate model, effectively capturing both seasonal variations and long-term trends. Holt-Winters struggled with the dataset's complexity, while LSTM performed better but did not match SARIMA's accuracy. These forecasts provide valuable insights for optimizing resources during peak periods and managing operations during low-demand phases. Future research could explore the use of hybrid models, further tuning LSTM, or predicting ridership at the station level instead of the entire network to enhance forecasting accuracy.